💬｜general-chat

[trail lion]

how could it, every image is different, even with controlnet there will be differences

[broken smelt]

right lmao

[astral goblet]

you need to use speicific controlnets to work with animatediff better

[trail lion]

they call it "temporal consistency" a fancy way to say the sh*t flickers

[verbal osprey]

What I'm talking about are the videos you see on reddit where someone is dancing, and they change the person to someone else or to make it a frog man dancing etc

[broken smelt]

tokyo_jab on reddit has a good method that essentially does the same thing but takes a different route with controlnet_tile

speaking of which, he threw up another video thats currently top when sorted by hot on the stable diffusion subreddit

[astral goblet]

theres sparse_ctrl rgb model. will let you give an initial image that it animates from

[broken smelt]

not a bad vid

[honest spear]

https://youtu.be/e1UgzSTicuY?si=cwuT9XYCO8fxZ23x

Here's an interview on why he left, it's very interesting.

[teal pagoda]

It's a long way until we'll have a good/decent AI video generator like SORA. Even SORA was overhyped (the devs recognized later that it won't even be available this year).

[astral goblet]

https://github.com/guoyww/AnimateDiff?tab=readme-ov-file#202312-animatediff-v3-and-sparsectrl

[teal pagoda]

best for now is AnimateDiff vid2vid, but if you have a good GPU, so you can have like 3-4 controlnets activated at the same time

[astral goblet]

while emad isn't ceo, he's majority shareholder still. and he's the one directing how stability is going to amalgamate with OTOY's rendernet blockchain

[trail lion]

yah, no way we get smooth consistent video now, even 2s worth (for realistic photos, I've seen lowres ones that look smooth)

[verbal osprey]

Ok...maybe I'll wait to dive into this in the future when it's much easier

[broken smelt]

emad stepping down and those researchers leaving might be the general fanbase's first time experiencing leadership changes. saw a ton of people who didnt even know who satya is lol. its a regular thing that happens once you have a career going for you -- always move onto bigger and better things. my company's got a ton of top c-suite moves in the past few years

[opal hedge]

According to the stand in CEO of SAI, 3-5 weeks

[trail lion]

yes and no, CEO changes, where the CEO is a major shareholder, tell me that's routine

like convince me that happens often

[opal hedge]

We have video with SDV, but it's like 2 second stills with a mostly stationary subject

[verbal osprey]

That would be awesome. What I'm most excited for is better prompt understanding and making more complex prompts.

[broken smelt]

hmm got me on ceo but at least at my place, cto and cfo did. id state more but thats more info about me than id like to share online lol

[opal hedge]

I believe this was the case with Runescape and JAGEX

Paul Gower was the CEO until he sold the company and then we got some dude who buried the game in microtransactions

[trail lion]

to me, it's usually just financially motivated, ie the golden parachute...probably most examples will be acquisitions, but this isnt that, which makes it unique

I honestly believe this guy is an idealist, but I'm nobody but take that fwiw

[dusk canopy]

should i post some of my 1.5 loras

🤔

[hazy wind]

How long and good are the descriptions of the CogVLM model used by SAI for SD 3?

[astral goblet]

i've seen people pull off miracles with sparse control . its very finicky for me. #🎥｜animation message

[dusk canopy]

holy fucking shit

the lora is 2 FUCKING GIGABYTES

HOW THE FUCK DID THIS HAPPEN

WHAT

[opal hedge]

jesus man, the hell you put in there

[dusk canopy]

20 images

[broken smelt]

i think ive seen that issue before

[astral goblet]

you were in the dreambooth tab instead of "lora"

[opal hedge]

Ah, it might have been a dreambooth

[dusk canopy]

isnt dreambooth more better then lora

technically speaking

[astral goblet]

"technically" but i mean,effectively.... not really

[dusk canopy]

also it only took 17 mins

[opal hedge]

It's aight, but you might as well use ip adapter+ for faces and character recognition instead then

[dusk canopy]

normally a lora would take me 30 mins

[trail lion]

in what sense? if you mean in the quality of the images you can produce in certain scenarios, maybe...but is that offset by the flexibility, and the the extra resources to train and store it

personally I like being able to hop between models to achieve certain things

[crude notch]

dreambooth (or finetune) can obtain more details

[dusk canopy]

20 images 2 gbs

shit aint worth it

[trail lion]

ok, but so what, add the details later

[dusk canopy]

ye fair point

[trail lion]

like do a run with denoise .4 and another with .8, and you can change models with lora, I think it's no contest

[charred mesa]

so uh

4-6 weeks huh

[broken smelt]

idk anything about finetuning but when it comes to it, i look to the local legend himself
#🔧｜finetune message
dreambooth: OUT

[dusk canopy]

it took 17 minws

with 20 images

[charred mesa]

its about to become 3-5 weeks

[dusk canopy]

dreambooth was faster then lora

[charred mesa]

ooh

[trail lion]

generating a full checkpoint for the goal of a concept or likeness is completely foolish, in my opinion

[dusk canopy]

e

im suprised my gpu did that in 17 minutes

💀

explains the ram usage tho

it was at 11.7 for dreambooth

[crude notch]

was

[charred mesa]

oooh

[crude notch]

lora, loha, lokr and such are faster

[astral goblet]

yeah if dreambooth is faster than a lora for you, you've got your lora config all effed up

[astral goblet]

youtubers teach this. its a popular misconception. that high quality likenesses can ONLY come from dreambooth full checkpoint training

[dusk canopy]

768x768 res

[astral goblet]

where a properly refined lora with generalized knowledge and a good prompt will do fine and work across a variety of models

[trail lion]

so if you put them side by side, with the model it was trained on...the dreambooth will be slightly better (on the first gen...which is all the youtubers ever consider)

but consider subsequent refining img2img gens, you'll be more than fine with lora

[astral goblet]

dreambooth is really destructive to a base model too. its just a lot less of a flexible approach. and innefficient for one likeness. so much energy wasted

[crude notch]

dreambooth / finetuning is like a really big sponge

loras n such are smaller sponges

[trail lion]

consider there are models out there that add quality, or unique attributes that wouldnt be available in whatever model you picked to train your dreambooth

[crude notch]

if its something small that it needs to learn, why use giant sponge when small sponge work

[astral goblet]

lolwut? thats not how it works at all

[crude notch]

analogy

big sponge mean more trainable params

small sponge mean fewer trainable params

[charred mesa]

ah

interesting

[astral goblet]

yeah i get it now. you don't need billions of parameters for one likeness

[crude notch]

you only need like 50m params for a good lora

[trail lion]

exactly, and dreambooth clobbers them anyway

[crude notch]

or 16/32 dim

[teal pagoda]

Look at Jensen (the Nvidia's CEO). He's there for Nvidia for 20+ years while Emad left after max 1 year. A true CEO stays with that company and tries his best to make it a successful one, not running away like a rat.

[astral goblet]

i find network dropouts help loras a LOT. each step it'll drop 10% of the digital neurons.. or whatever you set it too. those won't be trained . so it's generalizing the knowledge a lot wider

[crude notch]

drop out is useful yea

[charred mesa]

bit off topic, but do you guys know if the smaller SD3 models could benefit from more pretraining like how LLMs benefit for more pretraining?

[astral goblet]

different goals. don't believe the one true scotsman idea applies here

[crude notch]

uhhh no

[charred mesa]

thank you

[crude notch]

the small models are basically done

since theres only so much you can do with 800m / 2b

[charred mesa]

ah I see

[crude notch]

6b/8b are still cooking

[charred mesa]

and of course just DPO training mostly

if that helps?

[astral goblet]

iirc the clip and text encoders are the pretraining

[charred mesa]

I wish we could get the smaller models right now, to play around with it, but they wouldn't like that 😔

[crude notch]

its being dpo'd as of now

[charred mesa]

damn

[astral goblet]

oh poor sd3. unless i guess its into that

[crude notch]

what

[astral goblet]

huh?

[crude notch]

huh?

[charred mesa]

lol

yoinked do you know if the 2B model is close to SDXL quality or not

since it's almost reaching 3.5B

[crude notch]

yep!

[teal pagoda]

and wasn't it possible to just "decentralize" the AI at stability? :)) he ran away for more money

[charred mesa]

wow damn

that's good news

[crude notch]

from the one (1) example its doing better than 1.0

[astral goblet]

hes still majority owner of stability last i heard

[primal plank]

Is 3050 any good for SD3?

[charred mesa]

maybe highresfix with 2B would be good enough for low end users

[crude notch]

how many vram...

[charred mesa]

is it 6GB?

[primal plank]

8GB

[charred mesa]

ahhh

[crude notch]

i cant say but start uninstalling auto1111

and install swarm

[primal plank]

I have comfy

[crude notch]

great!

[charred mesa]

nice

[primal plank]

[crude notch]

also how much sysram

[opal hedge]

comfyui is bae

[primal plank]

16GB ram

[crude notch]

perfect

[charred mesa]

based? yes

[primal plank]

thanks

[charred mesa]

lol

16GB sounds good

[opal hedge]

That too

[teal pagoda]

I predict that by mid 2024 and full 2025, no one can do anything without a RTX 4090 minimum (I mean using the latest models and the best ones).

[astral goblet]

comfyui is the only ui with a fennec fox girl mascot

just sayin

[charred mesa]

I just wonder how much difference is between 6B and 8B, cause they sounds way closer to eachother than 2B and 8B

[opal hedge]

Nah, I think our models that already exist are really good and are only going to get more optimized

Honestly, future AI is probably going to be way less computation heavy imo

[charred mesa]

you mean like 24GB of vram is gonna be required for everythign?

[teal pagoda]

Not really, they'll need more features, more parameters.

[astral goblet]

in the year 20-25. man will need a 4090 too survive. 🎶

[teal pagoda]

Yea, minimum

[charred mesa]

I mean if we actually do get more VRAM lol

[opal hedge]

Yes, hopefully we learn how to use them well

[charred mesa]

nvidia is stingy with their consumer VRAM

[primal plank]

4090 is like 6 month of my salary :)))

[charred mesa]

yeahhhh

[crude notch]

lol

no

[astral goblet]

its more about the silicon forges not putting out denser chips for a couple years yet. nvidia doesn't want to compete against their richer markets

[primal plank]

Hope 50 series have more base rams

[crude notch]

36gb on 5090

[teal pagoda]

I believe you, man, but look at how fast are all the AI projects evolving

[charred mesa]

for inference? I believe the disagreement

[teal pagoda]

we'll see

[crude notch]

inference? nah

[charred mesa]

training too?

[crude notch]

train? yea

[charred mesa]

oh I mean like

[astral goblet]

tmsc won't be creating denser chips until sometime in 2025, and those won't hti products probably till december or 2026

[charred mesa]

inference won't need 24GB everywhere, but training probably would

[astral goblet]

they're still behind schedule due to.. well you remember the big ol shutdown the whole world had

[charred mesa]

I'll be lucky if I can train SD3 8B with 24GB with a lora

at decent quality

[primal plank]

can they reduce model comlpexity by using int8 or so? most of the SD models are fp32 or fp16 I think

[crude notch]

nope!

[astral goblet]

think we'll see models in fp4 or fp6? i wonder if

[crude notch]

or not int8

we could see some fp8 storage

afaik

[astral goblet]

i use fp8 everyday right now

[charred mesa]

but fp16 is still there for us

[teal pagoda]

not probably, 100% will. Even a dreambooth trained model for XL requires 48 GB VRAM

[broken smelt]

im fairly certain there are at least a few fortune500 companies with relatively new CEOs. seniority means little, it's about what you do in the time you have as a leader. the peter principle applies to the position as well -- once you've personally done the best you can in a management position then the best move sometimes is to step down and let new leadership try other things. we have so many instances where the opposite happens and people blame a company's "downfall" or turning point due to leadership not stepping down or "running it to the ground".

that's not to mention that the CEO is also a human. noone is expected to be tied down to their workplace forever lol. the previous CEO at my company has been doing amazing things last we checked in with him and none of that would've been possible if he had stayed behind.

[primal plank]

[charred mesa]

oh training yes, sorry

training will require a lot of vram

[crude notch]

lora for XL takes 12gb vram

[charred mesa]

well loras it's the "probably"

I hope I can train SD3 loras at 24GB

or we can get zero-shot type of stuff like IPAdapter

[trail lion]

and I must say again, this isnt some guy that was hired on during a transition, or interim, this guy was a major shareholder, invested, it matters

[broken smelt]

he was a cofounder, yes
is it relevant to him making meaningful decisions compared to if he stepped down and let someone else take over? idk, i dont know their financials or projects behind the scenes so i cant say

[teal pagoda]

Even these will require min 12 GB VRAM. Look at Instant ID

[charred mesa]

oh yeah

ipadapter with SDXL runs fine under 12GB of vram, but with SD3 8B, yeeesh

[teal pagoda]

Combine 12 GB with the GB needed to run the base model

[astral goblet]

you know what i like about jensen from nvidia? he's got the same name as the actor who plays dean and soldier boy

[charred mesa]

I wonder if comfy's mem management I can even get it at all

[broken smelt]

SAMMY

[teal pagoda]

you'll get around 20 GB VRAM needed

[astral goblet]

team dean all the way. sams a poser

[broken smelt]

i think of dean everytime i see jensen's name too lmao

the real team is team cas

[astral goblet]

castiel was best as cult leader cas

deanmon was peak SN

i been posting some sparse ctrl tests in #🎥｜animation i really think a lot of people are missing out on it. it's a gooder model

[teal pagoda]

The problem is that with this bad state of the world (and getting worse every day - prices at food, rent, gas, electronics go up, the salary go down or it's the same), it will be really hard for normal people who don't win any money from generative art to really buy a 4090 or the ones from 5000 series.

[broken smelt]

it looks like you perfected it in the span of 30 minutes lmao

the before and after are so different

[charred mesa]

well im getting a 3090 so its gonna be okay

but I will still feel bad for the 12GB users

[teal pagoda]

So we'll see that people will still use 1.5 models even when the SD3 will be out.

cuz they don't have the requirements

same applied with XL

[charred mesa]

well idk if they successfuly train the 800M or 2B models to corn with detailed prompts then they might switch

they will have controlnet, and run them at relatively low end requirements

[astral goblet]

what about me i hahve 16gb

feels bad too?

[charred mesa]

nah 16GB is good

[astral goblet]

feels good

[charred mesa]

its gonna be 100% fine for 16GB

[teal pagoda]

The problem with this "AI generative" hobby is that it's really "expensive" if you wish to be an up-to-date player on this field

[charred mesa]

SD3 8B, T5, with controlnets and other stuff

probably gonna run at 16GB

[astral goblet]

all hobbies worth the time are expensive imo

[charred mesa]

I just do these images for fun

so idk

maybe the 2B would have been fine all along

[astral goblet]

before i broke my hand a bunch, i was going to get into rock climbing. might still. but it's going to cost A LOT jsut to get some basic equipment

[charred mesa]

I just don't know how much worse it could be

[charred mesa]

yeahh

[teal pagoda]

not 100% of them, but most yea

[astral goblet]

maybe i wanna start canoing or kayaking. gotta buy all that gear. gotta buy a rack for my car. gotta have a car

[broken smelt]

my brother would probably wish his golf equipment cost as much as my pc lmao

[astral goblet]

lol golf yeh!

shit even hiking. gotta pay for gas to get to the places. and hiking in shitty gear.. oh man. sucks so hard.

[teal pagoda]

golf seems so boring for me

:))

no hate

[astral goblet]

wanna do archery? gonna have a cheapo fischer price bow if you don't invest

there's always collecting too.. but wowwweeee... pricey just to look at things

[teal pagoda]

I see your point, but you can use that equipment for some years, while in the AI domain everything is too fast

everything is rapidly evolving here

every single day I find new models and tools

[astral goblet]

new card every 3-4 years. thats not that much

i've been fronting that cost all my life basically

[teal pagoda]

nah, it will be needed once a year or at maximum 2

in the future

[verbal osprey]

Nvidia....give me 64gb of video ram in the rtx 5000 series....please....😆

[astral goblet]

doubt

in the future intel cpus will have inference support on die and it won't be a niche field anymore

[broken smelt]

me and the friends interested in AI all have 3090/4090s simply due to being gamers lol, the overlap is really nice

[crude notch]

impossible

[teal pagoda]

Nah, the consumers' cards doesn't have this

only the business ones

[astral goblet]

nvidia introduces the quadro RTX 5000. $50000

[verbal osprey]

I know it will never happen....I think we'd be lucky if we got 32

[crude notch]

(on consumer cards)

[crude notch]

36 on 5090

[astral goblet]

i know consumer cards won't get over 24 on the 50 series. tmsc put out their road map to higher density and it's not in the 50 series launch

maybe the 5090ti will see it

[verbal osprey]

That's too bad.

[astral goblet]

we're still in the period of manufacturers spiraling their supply plans because of covid complications

[teal pagoda]

You can get A100 with 80 GB of VRAM for 10K$

:))

[astral goblet]

the silicon shortage is real

[teal pagoda]

Yep, and not only that unfortunately

[astral goblet]

its always down to supply and demand

[teal pagoda]

and everything because of these apes with their covid pandemy lockdown joke and now with the restrictions against Russia + the useless joke "green deal" agenda

man, these apes don't have better things to do with their life

[astral goblet]

i wonder if nvidia are training a model to optimize chip architecture and manufacturing yields. that'd be wild

[astral goblet]

hmm. think i'll retreat into this hedge here after such comments

[teal pagoda]

Normally, right now a 4090 should have costed max 1000$ (in normal world conditions). But as the stock is so limited, they are still at 2000$+ unfortunately.

[teal pagoda]

this is the truth, man

believe it or not

I researched like crazy to find out why everything is going so wrong in the world nowadays

I'm not talking about conspiracy theories

The whole planet is affected by these decisions

the prices at food/electricity/gas in my country have doubled/tripled in the past 2-3 years

while the monthly salary is still the same or even lower

[astral goblet]

https://tenor.com/view/hiding-homer-bushes-im-out-gif-14300622

[honest mica]

I think this is just called corporate greed.

[teal pagoda]

Not all of it, but 60% yes

the rest of 40% is called government's greed/power

[teal pagoda]

you can't hide forever anyway. Not if a new war will start (hope it won't happen).

[forest trout]

Greed sort of undercuts what's actually going on. Companies have to grow or they will fail or get absorbed by companies that can grow faster. Greed actually doesn't play as big of a part as you'd think. It's sort of baked into the system.

[teal pagoda]

mostly the leaders of these companies ruin this world with their actions

they want more and more

[opal hedge]

Comapnies have to grow because the CEO is legally obligated to make it grow or else the investors can sue them

Even if it comes at the cost of long-term growth

[teal pagoda]

I know what you're saying, but not to start wars only because your weapons companies have new equipment and don't have new buyers for it

:))

so you have to use the old equipment/weapons to consume them, so you can buy the new ones

[forest trout]

You can't have long term growth if you go bankrupt first.

[opal hedge]

Usually old equipment and weapons are decommissioned if you're talking about lockheed martin and other government contractors

No one is going to war just to use weapons lmao

[teal pagoda]

nope, that's only one reason from the multiple ones

like the pharma won big sums of money in the pandemy, now the weapons companies are winning billions per year

[astral goblet]

"i'm not talking about conspiracy theories!" /describes conspiracy theories

[teal pagoda]

trust me, I made all the calculations

[opal hedge]

Yeah this is tin foil hat territory

[astral goblet]

crazy research indeed

[teal pagoda]

:)) if you think these are conspiracy theories, then it's clear

[astral goblet]

i must be one of ---- THEM ---- gasp

[teal pagoda]

just one thing out of curiosity. Biden or Trump?

[opal hedge]

Weapons companies always won billions of dollars per year, because we don't want to lose global hegemony (in the case of the US)

[astral goblet]

NDP but i want them to get a new leader

[teal pagoda]

you know, every single sane person with at least 2 neurons thinks that only a president like Trump can "get America out" from its misery atm

even foreign africans, asians and so on

[astral goblet]

oh

k.

[broken smelt]

im all out of tinfoil now

[astral goblet]

i'm all out of tinfoil i'm so lost without it i cannot protect my head from mind control raaaays

[teal pagoda]

I don't support anyone. I'm just with the truth. Even if you don't have 2 perfect presidents, you'll choose the little evil like it's said.

[honest mica]

Must be because of the current tinfoil prices. And everything because of the government.

[teal pagoda]

:))

exactly + World Economic Forum

gl hf

[astral goblet]

/buys cutting edge videocard and pays for broadband internet

WHY AM I OUT OF MONEY!?

[fervent thunder]

cheddar PACKAnerbs

[astral goblet]

must be the gummint. i can't be held responsible

[teal pagoda]

If you're from the European Union and you don't know what are the "reforms" from the European Parliament, it's sad

they are even publicly available

but it's a lot easier to not care about what happens around you

"it won't happen to me, idc" - we'll see in a good day

[true canopy]

finaly got my new computer set up, trying sdxl for the first time = )

[teal pagoda]

but enough with the "conspiracy theories", good luck

[rapid fulcrum]

https://tenor.com/TW2f.gif

[sterile raven]

What hashtags do you guys see the most success with AI art on instagram?

[astral goblet]

# girls

[karmic cedar]

It is happening. All you have to do is go and read the political lines, then read between them. Most people don’t like reading.

https://github.com/jasonppy/VoiceCraft

This is the other new audio model that is supposedly the bee’s knees or at least its ankles

[sour needle]

Can you use stable diffusion on phone

[low moon]

So what's ya'lls favorite checkpoints? Epicphotogasm is my fav 1.5 and wildcardXLturbo for sdxl... i guess

[karmic cedar]

Yes, look into the Draw Things app

I like Juggernaut XL v9 for photorealism.

As far as SDXL goes

[sour needle]

Draw wat

[mental bobcat]

ERROR: Could not find a version that satisfies the requirement torch==2.1.2 (from versions: 2.2.0, 2.2.0+cu121, 2.2.1, 2.2.1+cu121, 2.2.2, 2.2.2+cu121) ERROR: No matching distribution found for torch==2.1.2

anyone have any ideas on how to fix this error?

Im using an AMD radeon gpu if that has any relevence and ive tried putting --reinstall-torch as a command line argument

[broken smelt]

id post in #🤝｜tech-support, the more advanced folk hang out in there

[mental bobcat]

oki doki, thank you

[sterile raven]

Not sure if this was a joke but i think you might be right

[opal hedge]

Open source?!

[astral goblet]

half joking half honest opinion

[broken smelt]

the truth is a joke so thats fair

[fervent thunder]

supply chain attack on openssh by xz developer, ssh backdoor. fedora 40, debian testing/unstable/experimental, arch linux affected. linux users who have recently updated or installed distros should check their systems asap

cve-2024-3094

[astral goblet]

there's a few pretty serious cve's for linux landing lately

just finished this guy's tear down of one https://youtu.be/ixn5OygxBY4

[split kestrel]

when doing animations... what causes the radical shift in images? the motion scale? the steps? the CFG? or the batch size?

steps are 45 / CFG is 12 / scale is 0.8 / batch is 32

or is that just the nature of .... uhh sd

lol

[dusky relic]

I encounter a question:when i enable Caption option in Extra page when use stable diffusion webui,the command windows display 'load checkpoint from /root/autodl-tmp/stable-diffusion-webui/models/BLIP/model_base_caption_capfilt_large.pth' but when downloading finished,it pause,no any result came out...

[placid hatch]

Are there any good upscalers for sdxl?

[astral goblet]

context batch. each context window is it's own animation and they sort of just overlap and blend into each other

controlnets and loras help consistency. thats why the best animations are other videos with provided frames for guidance

over training a lora so that it is always one person in one outfit helps

not going to ask what discord servers you're scraping at all where too much makeup is applied. as that's often a thing teens do. sooo yeah. best to just leave those questions at the door

[fervent thunder]

the too poor paradox the too poor paradox theories that sentient agi wont become a reality as the knolage is held by someone too poor to actually create agi or that is beyond our current understanding of coomputers and physics =canceld

[astral goblet]

that poor person should git gud then.

fucker is holding us all back with his lack of skills

[fervent thunder]

lol

[dim minnow]

ai images

[fervent thunder]

also could be a factor of how our society views knolage by paper gains like money and degrees when my iq is higher than someone with a degree and went to private school

the world needs to change

for agi to be real

[astral goblet]

iq has long been known to not be an effective measure of knowledge

[karmic cedar]

Consider that many people involved in AI development hold philosophies of self that contradict the very neuroscientific principles underlying AI.

[fervent thunder]

the world doesn't view iq

[astral goblet]

1. you can't even spell knoledge. once was ok.. typo. twice though...

[fervent thunder]

typing with one finger

[astral goblet]

2. stephen hawkin's quote about iq is hilarious

3. you can't even home row

[fervent thunder]

bro went to epsiten island

how is he valid

[astral goblet]

4. you say shit like that

[fervent thunder]

generic name with wolf

[stone latch]

Ad Hominem

I have no idea what's even being discussed but this is the logical fallacy here

[astral goblet]

there are also different classes of iq tests. ones that are given to teens because their intelligence is often expected to be much lower, ones that are sold to adults who want to flex, and then there are ones that are designed for actual real geniuses. The intention of an iq test is a personal measuring stick. you'd do them over time to see if you're improving or not

[fervent thunder]

do you know what the founder theory is

its how Amish people develop extra thumbs and weird shit

over 100k years that would go neural

[astral goblet]

its also what race supremacists go on about all the time

[fervent thunder]

hence autism and brain deformities

the world is imbred

[astral goblet]

oh ok. /backs into the hedge again

[fervent thunder]

thats the biggest leak you will ever here

snowden is a drop in the bucket

[astral goblet]

https://tenor.com/view/wtf-what-bushes-gone-gif-14325166

[fervent thunder]

open your eyeys son

there are harard papers that touch on this but I expanded on it

harvard

truth hurts

[karmic cedar]

Kierkegaard said that talkativeness is the result of doing away with the vital distinction between talking and keeping silent

Oooh, I’m getting away free with that 😌

[shell tendon]

yeah, iq tests are ironically pretty stupid

[karmic cedar]

They measure an objective domain based on subjective criteria

Or rather, the opposite

I’m tired

[shell tendon]

yeah... just a pretty narrow assessemnt of a pretty narrow range of skills that say little about someone's real ability to make the complex simple, which i think is real intelligence

[karmic cedar]

There’s also the ability to interpolate ideas, aka creativity

[shell tendon]

grades are pretty weak on that too... if you have really good grades, sure, it says something, but it's far from definitive and still too narrow

yeah for sure

[karmic cedar]

And grades are again based on an objective criteria

[shell tendon]

theoretically

typically applied subjectively

[karmic cedar]

I think in general, if complexity can be demonstrated with context, it is acknowledged as intelligent

[shell tendon]

and typcilaly only assess one's ability to solve a problem that is at worst closely related to one that's already been seen and worked through

[astral goblet]

like any personal assesment test, they're useful as a personal development tool

using an online internet iq test to believe you're smarter than stephen hawking, might mean you aren't well developed

[shell tendon]

my take on basically everything is... ignore all the BS crap unless you're required to do it, and just focus 100% on getting shit done

lol yeah

[karmic cedar]

In the end it’s what you believe, right?

lol

[pearl ocean]

theres tons of Models on civitai, but they are all based off SD, so they all kinda look the same.

so I guess that makes sense

[karmic cedar]

what is the worst album cover

there must be one

[shell tendon]

lol

[karmic cedar]

i went with www.thug.com by trick daddy

(using Become Image)

[lean rapids]

someone can help me with a problem at the moment of download a model

[pearl ocean]

the one you don't know about

[karmic cedar]

😦

reminds me of philosophy class

(long time ago :-)

[jade tide]

What is the best model for interior image generation like the stuff you can find here? https://interiorai.com/interior-designs/vintage-style-living-room-interior-with-occasional-tables-and-chairs-and-rug-a-15629

[karmic cedar]

the one you don’t know about

[fallen drum]

3090 ti 24 gb vs 4080 16 g ?

[buoyant marsh]

[teal pagoda]

Exactly. Tell this to them. They call everything a "conspiracy theory" :)))

[teal pagoda]

You shouldn't have used that argument

[teal pagoda]

3090 or save a little for the 4090

[amber bloom]

or save a lot for the upcoming 5090 😉

we have time

[static cape]

Any news or new preview images?

[loud solar]

😁

[opal hedge]

Man, apparently SAI is in deep water financially

[amber bloom]

They need to stay solvent long enough to release SD3 to the public

After that I expect some big corp will buy them out, probably for cheap

[teal pagoda]

Guess emad already knew that and got out of this

:))

[rugged mirage]

it doesnt matter if they test a narrow or wide domain, what matters is that they correlate fairly well with a bunch of other things which they do

[violet fog]

Hey, guys!

I'm new here, and I was wondering if I could use this discord server to generate images with SD

[opal hedge]

I hope this happens so bad

[opal hedge]

No, it's been disabled

[fervent thunder]

Hey i need some help i cant figure it out like rarely anyone talks about it but, I've tried every webui and when i switch to ponydiffusionXL or any XL it just crashes might anyone know the reason why?

[trim magnet]

not enough memory

[fervent thunder]

im sorry 64G Ram is not enough

it was literally working 3 days ago

and it just shat itself

when i swap to any xl model i see my ram getting flooded

[trim magnet]

if u are using the normal a1111 thats probably why,try with forge

[fervent thunder]

i did that

worked for 1 day

and it stopped working

i tried invokeai took me years to download just for it to hit me with a crash

so the only webui works is comfyui

in the meantime if anyone finds a fix please tell me i dont want to use this dogshit webui

[trim magnet]

try to ask in tech support channel,upload a screenshot there of your cmd window to see if u get any errors

[rugged mirage]

SD3 is pretty much done, people are already testing it, so Id imagine that even if they ran out of money next week we'll still get a release

[white finch]

hello, i need someone to help me install stable difusion to my computer 🙂

can anyone please help me

[rugged mirage]

follow a tutorial from youtube

[white finch]

the problem is my internet speed is slow, when i trying to install stable difusion

[rugged mirage]

there isnt a minimum internet speed you need to install it

[white finch]

with cmd

it broke and not install

Microsoft Windows [Version 10.0.22631.3374]
(c) Microsoft Corporation. All rights reserved.

C:\A1111\A1>git clone https://github.com/vladmandic/automatic
Cloning into 'automatic'...
remote: Enumerating objects: 43156, done.
remote: Counting objects: 100% (260/260), done.
remote: Compressing objects: 100% (135/135), done.
error: RPC failed; curl 92 HTTP/2 stream 5 was not closed cleanly: CANCEL (err 8)
error: 6766 bytes of body are still expected
fetch-pack: unexpected disconnect while reading sideband packet
fatal: early EOF
fatal: fetch-pack: invalid index-pack output

C:\A1111\A1>

Please help me .

[rugged mirage]

1. use #🤝｜tech-support for tech support not a random other channel 2. just delete and try again, seems like your internet just cutoff or something

[fervent thunder]

god forbid a man have fun using an sdxl but then is forced to a shitty webui

[regal snow]

It sounds like you are teetering on the edge of the minimum requirements… Forge is better at managing resources vs A1111.

[trim magnet]

i run it perfectly on forge and only have 16gb of ram,sounds like you have a gpu issue

[opal hedge]

You need vram, not ram

[fervent thunder]

bro

explain this

it was working perfectly yesterday

i wake up today god knows why

[charred mesa]

bruh

[fervent thunder]

it just started shitting itself

[opal hedge]

How is it crashing, describe the exact process

[fervent thunder]

Loading weights [ac006fdd7e] from J:\New folder\Terminate\2005463692\Content\Items\yup\Aib\webui\models\Stable-diffusion\autismmixSDXL_autismmixConfetti.safetensors
model_type EPS
UNet ADM Dimension 2816
Using xformers attention in VAE
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
Using xformers attention in VAE

i load it

load an xl model

[trim magnet]

then u updated something,installed something

[fervent thunder]

wait for it

crash

[fervent thunder]

probably

[opal hedge]

Try comfyui

Does it work?

[fervent thunder]

it works but its comfyui

i despite it

[opal hedge]

So it's an A1111 issue

[fervent thunder]

yep

[opal hedge]

Try a fresh install

[fervent thunder]

i tried invokeai

crashed aswell

when i did a fresh install it actually worked

so is there no other possible way

other than just fresh installing

[opal hedge]

Yep, start fresh is all I can say

[fervent thunder]

alright

this reminds me of fucking cyberpunk

everytime i setup my mods

surprise update

fucks my mods

[opal hedge]

Turn off auto updates

In both cases

[fervent thunder]

i did that for cyberpunk

still is backstabbing me

[charred mesa]

back in my day we had A1111 get fucked up from auto updates every week and we didn't complain! 👴

[fervent thunder]

how do you turn off autoupdates?

for a1111

[fervent thunder]

okay old man

[charred mesa]

show the .bat file

[opal hedge]

Idk, I'm a comfyui gigachad

[charred mesa]

same fr fr

[opal hedge]

I don't use peasant software

[charred mesa]

lol

show the .bat file

which you start it with

[fervent thunder]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --no-half-vae

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
@REM set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
@REM --ckpt-dir %A1111_HOME%/models/Stable-diffusion ^
@REM --hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
@REM --embeddings-dir %A1111_HOME%/embeddings ^
@REM --lora-dir %A1111_HOME%/models/Lora

call webui.bat

[rugged mirage]

A111 is just such a nicer UI and easier to figure stuff out if you dont live and breath SD, Comfy is only better if you are constantly in it and spending time on it

[fervent thunder]

🤫 🧏‍♂️

[opal hedge]

a1111 ui confused me, comfy seemed so much simpler

It was just too much junk everywhere

[fervent thunder]

started spitting facts🗣️

[fervent thunder]

bro what

comfyui is for masochists i dont know what to tell you

[rugged mirage]

a111 you can see the basics and then just read the options in the menus, in comfy you have to make workflows out of thin air

[fervent thunder]

he might be illiterate

[opal hedge]

Comfyui is really neat and simple once you know how it works, A1111 has a million tabs that you have to shuffle through

Different strokes for different folks I suppose

[charred mesa]

I started out with A1111 myself of course, then used comfy cause I heard it runs SDXL better, then stuck with it ever since lol

I could be using forge with similar speeds but I just use comfy anyway

[rugged mirage]

Im moving to Comfy now because stuff usually come a bit earlier, and if you put in the time you have slightly more options but if you dont use it all the time it's awful

I can leave A111 for 3 months come back and figure it out, in Comfy idk what fucking nodes to even search for for every little thing unless im already deep in it

[charred mesa]

Comfy because stuff usually come a bit earlier
This 🔥

[opal hedge]

Comfy lets you do crazy workflows

A1111 seems too handholdy imo

[charred mesa]

but its recommended for beginners

[fervent thunder]

An idiot admires complexity, a genius admires simplicity - King terry davis

[opal hedge]

As quoted by a man who lost his shit moments ago over a program that he was 'backstabbed' by that he now claims is simple

We live in a society

[honest spear]

I'd say you only see the value of comfy when you have a very specific goal or workflow in mind, and you realize that it's 100% possible

[fervent thunder]

As quoted by a man saying a1111 is "handholdy" after saying that a1111 ui confused him with his challenging illiteracy

We truly live in a society

[kind geyser]

I finally managed to run stable diffusion on my geforce 940m

[karmic cedar]

George Lucas uses A1111

[fallen monolith]

Yoo can someone tell me if I can run automatic1111 of an intel iris xe graphics GPU?

[fervent thunder]

That's because he is based

[karmic cedar]

WHAT George Lucas????

[vivid lance]

comfys preformance is so so much better

I do prefer A1111's layout but yeah its just not worth it

[fervent thunder]

That's the only thing i like about comfy

I can play games while running comfy in the background generating images

[vivid lance]

with the impact plugin and my presets built I feel I am not missing anything from lack of automatics

[fallen monolith]

Come on gang don’t leave me on read like that

[opal hedge]

A1111 is good for people who'd prefer ease of access over depth and functionality

[fervent thunder]

who let him cook?

[vivid lance]

I have had more problems using automatics too over the past 5 months or so lots of bugs sending img-img etc

[opal hedge]

Does forge give the same problems?

[vivid lance]

got no clue never used forge

[fervent thunder]

Problem is i do have to switch to comfy sooner or later

a1111 has to many dogshit bugs

including forge

[opal hedge]

Very likely not, I doubt it has CUDA support

Also, no VRAM means it won't work

[fallen monolith]

Ahhh okay well is there any alternative I could use that you know of?

[opal hedge]

For image generation or an alternate gpu?

[fallen monolith]

Image2image generation

[opal hedge]

I think midjourney does img2img, leonardo.ai might as well

You can also try stable horde

Or at least look into it, I'm not sure if your gpu would work with it

[fallen monolith]

Okay I’ll have a look and see it works

Appreciate the help broski

[opal hedge]

np, hope you find something that works

[rocky island]

hi everyone. which better 4070 or 3080ti?

[fervent thunder]

3080ti>>>

[charred mesa]

the 4070 has better power efficiency and av1 encoding support (videos/streaming look better), but the 3080ti just was waaaaay more raw power

[woeful forum]

Hi there! Stable Diffusion newbie here with a question about LoRA-LCM. I use AUTOMATIC1111 and the SD 1.5 model (not the XL one).
I've come across two tutorials on LoRA-LCM - one (https://education.civitai.com/quickstart-guide-to-lcm-lora-acceleration-modules/) tells me to download this file: https://civitai.com/models/195519?modelVersionId=225222
while the second one (https://stable-diffusion-art.com/lcm-lora/) says I should get this one: https://huggingface.co/latent-consistency/lcm-lora-sdv1-5/blob/main/pytorch_lora_weights.safetensors

I followed the first tutorial and downloaded the suggested LoRA file. However, the results I'm getting in AUTOMATIC1111 don't match the ones shown on the civitai website. Now I'm not sure if this is because there is something wrong with my LCM, or maybe it's the models I have. To isolate the issues, I'd like to start by ensuring I have the correct LoRA file.

So which LoRA file should I use? Does it even matter which one I choose?

[charred mesa]

it has 384bit memory bandwidth

[fervent thunder]

well said 🙏

[charred mesa]

don't know, but you can 100% trust the huggingface one

[opal hedge]

Neither get a 3090 for the 24GB vram

Otherwise they're almost the same thing except 3080ti eats power

[rocky island]

ok thanks

[woeful forum]

Thanks! BTW, they say "First, download the LCM-LoRA for SD 1.5 and put it to the LoRA folder stable-diffusion-webui > models > Lora. Rename it to lcm_lora_sd15.safetensors." - just curious, why do they want me the rename that file, does it matter? Or is this only for my convenience/readability or what?

[charred mesa]

If you are using A1111 or Forge then that may actually matter. If you are using comfy them not as much I think

[woeful forum]

OK, thanks! I hope this will work 🙂

[charred mesa]

good luck

[worthy bone]

hey guys so no one got accepted to try SD3 early? I guess this waitlist is useless 😂

[trail lion]

people are getting accepted, it's just a small number, model trainers, etc. the rest get to be patient

[charred mesa]

and private partners and researchers

just look around and you might find some that accept prompts

I got lucky with that twice

[broken smelt]

Yeah I've seen some folks take requests recently in some reddit comments

[vale steeple]

They have been sharing around prompts there, yeah

[opal hedge]

There are a few people who were accepted. Some of them were taking prompt requests on reddit earlier today

[broken smelt]

I think if you really really wanted to try out a prompt, you could beg the Twitter accounts of the ones publicly showing their sd3 results, I'm sure eventually one would give in

Heck even Emad was taking requests on Twitter last week

[charred mesa]

I asked emad like probably only 1-2 hours after he went to sleep and next day he didn't take requests loll

[hasty hornet]

people still using twatter?

[trim magnet]

yes,specially the ones who spend the whole day fighting over politics/celebrities/religion

[charred mesa]

I only made it to keep up with Lykon, Emad and others

but they all died down so uhh

might as well delete the account after SD3 comes out

[hasty hornet]

I don't have to, I got banned or restricted, whatever they called it, doesn't matter, I guess

oh wait no, it was other account...on new one they just threw me out and asked me my nickname, which I just don't know, because they added random numbers to it which I don't remember lol

[peak patrol]

Hello, I am new to stable diffusion,
I have a question about Image To Image, how to change the style of the original photo according to the prompt but still keep the face and pose the same?
Thanks a lot!

[exotic loom]

try to play with Denoising strength and Sampling steps, to make the img look more like an original

[peak patrol]

thank you for your suggestion, I tried a lot with the Denoising strength and Sampling steps parameters, but the results were not as expected.

[honest spear]

Use controlnet pose to keep the pose the same, for face controlnet or inpaint. Style use prompt or a style transfer model, I think even sdxl has one or two now

[peak patrol]

Thanks @honest spear

[opal hedge]

It's mostly bots at this point and a few people too slow to notice it yet

Dead internet theory strikes true

[opal hedge]

You can try clip skipping and see if that helps

controlnets too

You can do ipadapter for face and control net for pose

[peak patrol]

Thank you for your suggestion

[midnight yacht]

Bots are pretty funny.

[astral goblet]

this is why children need to be sequistored on their own internet and adults require a license to acces it

case in so many point

[sweet token]

First time I've seen such a troll on this server

[midnight yacht]

This is cute.

[astral goblet]

happens all the time. mods are employees though and stability is closed on weekends

[sweet token]

Well I haven't been on this server for long time so I don't know

[midnight yacht]

Interesting

Good Friday was pretty nice.

[astral goblet]

lol dudes got serious mommy issues

[charred mesa]

This guy hates brians

😔

[astral goblet]

his switch (non oled) probably got taken away

[charred mesa]

[sweet token]

So how's everyones day been so far?

[midnight yacht]

It's interesting Steam allowed AI art.

[sweet token]

This has to be considered spam

[opal hedge]

Just block him and call it a day

[sweet token]

When?, I haven't heard of this news

[astral goblet]

i just reported his account for hate orientated violence. if a few other people do it, his account will be suspended by monday morn

ez pz

[charred mesa]

naughty naughty words

[midnight yacht]

January, they allowed it. I saw some games already starting to release with AI-generated Backgrounds, and said artists were still involved still. It's interesting to see some artist embracing it as well instead of complaining. I personally think people who appreicate artist's works will still commission artists.

[sweet token]

Don't respond to him, you are only feeding him more

[charred mesa]

oh im just having fun

[astral goblet]

fun watching these X blocked message counters just spam upwards lol

its like i win points whenever it goes higher

[opal hedge]

Did anyone read today's forbes article on SAI?

[astral goblet]

winner winner chicken diner

[opal hedge]

https://www.forbes.com/sites/kenrickcai/2024/03/29/how-stability-ais-founder-tanked-his-billion-dollar-startup/?sh=43bbab053e63

[karmic cedar]

Systems, man.

[charred mesa]

lol

[sweet token]

Interesting article title

[astral goblet]

the only relevant news about stability will be about their inevitable acquisition. whoever ends up buying them

[midnight yacht]

I am not subscribed to it.

[opal hedge]

Gotta get that clickbait

[warm junco]

Give me a sec then hes gone

[astral goblet]

i'm hoping for hugging face since they publically fished it already

[opal hedge]

"Here's a summary for people who aren't reading all that. Note that I'm just condensing the article and not necessarily agreeing or disagreeing with anything - don't come arguing with me personally.

• Stability AI, once a buzzy AI startup valued at $1 billion, is floundering and on the brink of collapse due to overspending and mismanagement by its founder and former CEO Emad Mostaque.

• Mostaque made grandiose promises he couldn't deliver on, like building custom AI models for nations, and poured tens of millions into research without a sustainable business plan.

• By October 2023, Stability had less than $4 million left and mounting debt, including overdue payments to Amazon Web Services. It was projected to lose over $140 million in 2023.

• Mostaque's spending on expensive computing power to develop AI models like Stable Diffusion rapidly depleted Stability's funds. Efforts to monetize the technology through APIs and enterprise deals largely failed.

• After investors like Coatue and Lightspeed lost confidence, Mostaque was pushed out as CEO in late March 2024 after an attempted fundraising tour flopped.

• Stability is now being run by interim co-CEOs as it scrambles to find more funding or a buyer. Looming legal threats over copyright issues add further uncertainty about its future."

[charred mesa]

balls

🏀

[karmic cedar]

Someone hasn’t had their ADHD meds

[charred mesa]

normal pills?

[warm junco]

And hes gone 🙂

[astral goblet]

the usual story is their parents have their pills for them but the kid hates their mom so they dont take their meds.

[charred mesa]

thanks CS1o

[midnight yacht]

Hmmm, but they did make it free

[charred mesa]

I didn't even know that being ADHD suddenly causes you to become a neo-nazi and spam slurs

🤔

but okay then

[sweet token]

This is making me worried about the future of stability ai

[astral goblet]

its a tantrum. adhd causes temper tantrums

[charred mesa]

@spring lark already banned

[opal hedge]

I think the takeaway was that Emad wasted money. He was renting GPUs from Amazon instead of building his own serverbase, so the company has little physical assets to back up on

[spring lark]

crazy chat moment

[charred mesa]

yes

[opal hedge]

He also funded random projects that had little world benefit or use in AI

[charred mesa]

imagine having been here since 2022 sept 20th

[midnight yacht]

I see

[spring lark]

imagine having been doing SD since its early days

[charred mesa]

fr fr

[karmic cedar]

He went about it as an individual would.

[charred mesa]

we had terrible prompt adherence and we liked it! 👴

[opal hedge]

Early SD days was basically "omg, that almost looks like a face!"

[charred mesa]

we had spaghetti for each and every prompt yet we like it! you guys complain too much! 👴

[sweet token]

True

[astral goblet]

i'd take eveyrthing in that forbes article with a grain of salt. they're a click bait rag that tries to affect stock prices for investors. market manipulation skirting experts

[charred mesa]

hell yeah 2022 august 1st

[spring lark]

frfr

[spring lark]

back in ma day 🧓

[opal hedge]

Now we're like, "the titties aren't quite rotund enough"

[karmic cedar]

Some of the best glitch art

[sweet token]

I remember SD 1.4 like it was yesterday

[charred mesa]

same

Time goes fast when you're havin fun! 👴

[spring lark]

horny people on SD1.5 be like

[opal hedge]

It was last year lmao

[charred mesa]

LAST yeaR???

[astral goblet]

2

[sweet token]

I remember the hype when they opened the bot to early testers

[opal hedge]

Yeah, 2022 feels so far away

[midnight yacht]

Funny how it's only copyrighted images when things start looking pretty. But it was all right when it was generating four legs and and three arms

[astral goblet]

i started on stable diffusion between the 1.4 and 1.5 release. that was october 2022

[opal hedge]

Artists in 2022: AI art will never replace artists
Artists in 2023: AI art has no soul!
Artists in 2024: It's stealing our jerbs

[charred mesa]

I remember watching others waiting for DALLE2 access, seeing image generation actually being possible was thrilling.
Then I heard about this Stable Diffusion thing, downloaded it and was amazed. I could generate images OFFLINE! It was wonderful.

I think I got SD1.4 around august 30th

on my laptop

[spring lark]

how long did that thing take

[astral goblet]

again demonstrating how misinformed you are. studios that higher artists aren't about to stop hiring artists

[charred mesa]

rtx 3060 laptop

it was pretty good actually

[opal hedge]

They won't?

[spring lark]

oh
i thought some dell 2015 laptop

[charred mesa]

naaaah

[midnight yacht]

It's just inconsistent arguments, "the shadows are off, the details are off, it has no soul", but then "its stealing copyrighted materials and taking jobs". If it was so inferior, soulless, and looks like trash, why do you care?

[astral goblet]

just look at the ai art that non artists create and you'll understand

[charred mesa]

6 GB was just enough at the time

with lowvram and medvram

then Xformers came and everything got so much better!

[sweet token]

I remember my first use of stable diffusion when its a just a channel where you could request your prompt to be ran by mod

[spring lark]

were using comfyui or a1111 tho

[charred mesa]

faster speeds and no medvram/lowvram needed

[opal hedge]

I write books, and I used to hire artists to do concept art for me. Now I can do it all myself. It's nice not spending 300 dollars on a cover

[charred mesa]

I used something like fast-stable-diffusion or something

cause I wanted to see if I could run it on a measly 6 GB of VRAM

[spring lark]

goofy ah
isnt it abandoned by now

[charred mesa]

probably, yeah

[astral goblet]

big doubt on that

[charred mesa]

you can do art covers for those with AI?

without any legal trouble?

[spring lark]

this?
https://github.com/TheLastBen/fast-stable-diffusion

[charred mesa]

aren't images generated CC0

[charred mesa]

ehh not sure

Stable diffusion turned from a cool novelty into a photoreal portrait/corn generator 😭

[astral goblet]

authors don't need concept artists because they are writers. why would you conceptualize in that process?

individual cases don't matter either. the market for artists with actual artistic skill won't dry up. you're just not that important in the grand scheme

[opal hedge]

No, you don't have to pay royalties on individual images. But renting out Stable Diffusion or offering services with it costs money

[charred mesa]

no like with the old versions

[spring lark]

the horny people always get to AI with their goofy stuff

[charred mesa]

SDXL and below, with commercial openrail licenses

[astral goblet]

i think for a bit, stable diffusion affected only fans content creators. like there was a site wide dip. but then all the simps and the content artists realized, higher production value is wanted. market demands and it receives

[opal hedge]

You can make really high quality images with SD right now. I can show you some

[spring lark]

well as reading here
https://stability.ai/news/stable-diffusion-public-release
it says

i) The model is being released under a Creative ML OpenRAIL-M license. This is a permissive license that allows for commercial and non-commercial usage. This license is focused on the ethical and legal use of the model as your responsibility and must accompany any distribution of the model. It must also be made available to end users of the model in any service on it.
sooo
its prob CC0? or CC1 ig
idfk what model tho

[opal hedge]

It looks like concept art from a league of legends character splash screen

[charred mesa]

CC1?

[spring lark]

https://creativecommons.org/publicdomain/zero/1.0/deed.en

[astral goblet]

sure . you can cobble something together that "looks" good. but it's extremely limited

[spring lark]

idk last time i checked SDXL has high potential with high res fix and good workflows

[astral goblet]

so then you start learning about what makes good composition and design and how to target art for your intended audience, and then realize someone with expertise on these matters would be valuable

it sure won't be managers that are using generative art tools in the future

i was around when photoshop was a brand new tool that was causing a lot of controversy. Digital painting was outlandish and not "real art" . its all the same pony show over again

[honest spear]

The fact is that normal artists can do orders of magnitude better, but if the threshold required for a job or commission is low, all that potential is wasted or not required

[astral goblet]

even photographers were hating on photoshop and trying to explain all the reasons why dark rooms weren't going away anytime soon

[opal hedge]

I think big companies and more patron-friendly media will still use real art for a while, but there will come a time where ai art blows most real art of the water. I think it already does by a large margin. It might not be as good as the top 1% of artists, especially commercial artists, but I can already make stuff equal or better than 90% hobby artists.

[astral goblet]

there's already low effort projects that exist. asset flipping . license a huge pack of stock assets that artists made, and then assemble them into a very low effort product.

that market has a big giant cap on it

[stone latch]

Exactly. I think it's important to be open-minded. There are tools like Suno.ai that I find cool and even play around with myself, and yet I produce original (non-AI) music.

[astral goblet]

the barrier of entry for producing music is at a historical low. the skill floor for mastering audio is still pretty high though.

aanyone with an experienced ear can instantly hear amateur work

[honest spear]

audio is scary 😄

[stone latch]

That's right, and I can tell that Suno is definitely not perfect, but I'm still not completely against the tool

[astral goblet]

ask ai to craft music for a game level. that music isn't going to fit the level at all in most cases. so you learn about pacing and mood and realize that someone with expertise on these matters would have some value

[honest spear]

I see ai audio helping a lot experts, like the ones makings audio effects in Star Wars, some of them used anything to get some of those sounds, like a pig hitting a wall, like anything mixed together

[opal hedge]

Man, ai voices and chatgpt were huge for skyrim

Hopefully more games in the future have it so that NPCs have opinions on everything and infinite dialogue

It'll probably be indie devs who implement it first, but the future of gaming is looking bright

[honest spear]

there was a post on x from the producers of a game I think Avalon, ai npcs can also decide to not help you or not giving you a quest if one of your companions had some problems with them, the system is getting very refined

[astral goblet]

not really. they used existing voices from npc voice actors to create sex mods for them. it wasn't a great look for the mod scene

it blew up because of the controversy it generated. its only useful to people who want to do sex mods in skyrim though. a very niche community

[opal hedge]

I haven't use it for sex mods myself, but I have had conversations with innkeepers about the ongoing civil war situation in skyrim

[astral goblet]

if i didn't own a lifetime membership to nexus mods, i would've been one of the people cancelling their subscription due to their lazy enforcement of it

[astral goblet]

chat gpt enabled engines? meh. another pointless subscription cost. and those chatbots got that extra chromoson problem. cool if you're into that i guess but.. meh. not a big deal

i worry for the future where single player games will require an always online connection because of the NPCs. we sure wont be running an LLM while playing the game on the same gpu

worst game ai upgrade i've ever seen to be honest. turning dialog into an online service

[small lily]

Where did the bots go? how do we generate now?

[astral goblet]

https://developer.nvidia.com/ace games that use this will flop and development studios will realize that there is value to writers and voice actors.

[opal hedge]

You can do both. Writes and voice actors for core gameplay and cutscenes, and then AI to add longevity and life to side content

The problem is writers and voice actors can't script for a single game forever. You need bulk at some point, especially for live service

[astral goblet]

the Finals, good game, they hired voice actors and trained models on their recordings. so that they could do new lines without bringing the actors back in. thats going to be huge in the future

[opal hedge]

Yeah, that's pretty terrible

I think we need laws about that

And lawsuits big enough that the voice actors never have to work again

[astral goblet]

i think its' fine. voice actors got paid and negotiated a proper license for those recordings.

they can then train their voice for a new project immediately. less hours in a recording booht on one project is a huge gain for them. voice acting is grueling on the chords

i dont think the finals devs have a license to use their voice lines or derivative products on future products. just the one

[opal hedge]

If the voice actors agreed to it I think it's fine. In hollywood, some companies are cloning actors' appearances and voices so they never have to hire them again, and this was done without the actor's agreement

I thought it was that situation

[astral goblet]

the longest actor strike in my lifetime happened over that one. the union won

james earl jones noteable gave disney permission to train an ai of his voice , exclusively for starwars products.

[pliant jasper]

fine w/ this as long as the voice actors KNOW that this is happening, and there's terms on how the synthetic voices can be used in their contract

[astral goblet]

Yeah its a great tool for developers and performers to collaborate with! Especially if voice actors don't live locally. flying them back in for new recording sessions isn't always ideal

a lot of voice actors destroy their larynx while performing too. knowing they will have less recording sessions will allow them to go even harder in the booth. spittin that hot fire. mans not hot

mans never hot!

[woeful forum]

Hi. I don't get it - a simple prompt (House, campfire, people having fun, sunset, raw photo, 4k) and this is the result: https://i.imgur.com/nPF9Wiy.png - why does it look so horribly 😦 I'm a complete newbie btw. Negative prompt: FastNegativeV2, Steps: 50, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3080215138, Size: 512x512, Model: epicrealism_naturalSinRC1VAE.
I tried different models as well (like DreamShaper). Why the faces are so horribly deformed?

[opal hedge]

Because it's SD1.5 which is an old model. You need additional gizmos to make it work

Either start learning about facedetailers and upscalers or go for sdxl models

[woeful forum]

oh! So it's not about my prompt? OK, I'll have a better GPU soon, until then, what is this "facedetailers" about?

[opal hedge]

It makes faces look good-er

[woeful forum]

copilot and google tell me it's a comfy-ui thing (this facedetailers), I use AUTOMATIC1111 - I'm too stupid right now to get into this comfy-ui 🙂
Will this SDXL work on my CPU, without GPU? I don't care if this will take longer to generate an image, it's only about playing with Stable Diffusion (until I get a real GPU)

[trail lion]

Maybe, but if you have no GPU just use an online service, or rent GPU if you're concerned about privacy

[vivid lance]

why not both 😄

[woeful forum]

It's all about education, I know I can rent some GPU power in the cloud, but when it comes to how fast I learn I think it's faster on my local PC. Might be wrong though.

[trim nymph]

https://x.com/artsiom_s/status/1774125236234924173?s=46

sd 3 turbo

[charred mesa]

yeah there's a paper for it, its great

its like SDXL Lightning rather than SDXL Turbo

I might actually daily drive turbo if the prompt adherence still proves to be better than midjourney

[exotic loom]

Hello to everyone, I have a question about training stable diffusion webui, I have 2 videos first one one is normal and second one is the same but with cool 8 bit effect that I cant repeat in any editing software, the videos are from pxlprsche, as I understood he took the video and passed each frame through ai, I want to train my own ( I tried to find if he used already made one but didnt find any that makes this effect) so it gives me same effect, so to say I have a lot of his material to feed to ai but I dont if it possible to say to ai in stable diffusion " look I have this original img and here is the one with effect i want you to be trained so you can make the same effect to img i give" is it possible with stable diffussion or i need to use another software of ai?

[opal hedge]

Babe wake up, new research just droppped

Ah, nvm this is the old sd3 turbo paper

[charred mesa]

^

[karmic cedar]

lol “old”

[gaunt pulsar]

2 days ago is 2 years ago in SD3 hype time

[karmic cedar]

right

[broken smelt]

The singularity is just two research papers away!

[astral goblet]

what a time to be "alive"

[charred mesa]

Don't look at where we are, look at where we will be, two more papers down the line!

[astral goblet]

if you call dealing with skyrocketing cost of living at unprecedented rates "living"

[opal hedge]

Back in those days, Emad was still CEO and we had hope for the future

[karmic cedar]

Thank you Karoly

[astral goblet]

the fusion of kakarot and broly?

[karmic cedar]

Two Minute Papers, the ultimate YouTube AI science paper channel

or one of them at least

his signature saying is “What a time to be alive!”

also “Hold on to your papers!”

[astral goblet]

i like brolly more.

[karmic cedar]

broly = more brevity for sure

[hallow plaza]

Any advice on how to make your own linkedin profile pic with SD?

[karmic cedar]

Consider training a LORA off of SDXL.

That would probably be the most precision-focused way.

If it’s a professional network you’re using your photo in, expect scrutiny lol

[hallow plaza]

Anything specific for this use case?

[karmic cedar]

You could try spinning one off of a realism finetune…

Check out civitai

[astral goblet]

legwork? ughhh

[mortal delta]

Hey does anyone know of any cool or creative things that can be done with free ai that have some use or are just entertaining, Im asking because im bored but im trying to be creative, but i got like no ideas.
Ive also grown to want thing as soon as possible and i dont know how to resolve this.

[karmic cedar]

“Cumdetective is typing”

[karmic cedar]

That is a matrix of issues there. I’m sorry

[mortal delta]

i guess im complex like that.

[kindred vigil]

what are all of you guys's gpu, and how does stable diffusion run for u

[karmic cedar]

Creativity for me is like a gas, it just sort of floats in and out

Like a fart

[astral goblet]

ahh. creative blocks. the tormented artist's mainsale

[mortal delta]

amd 480, it runs not great.

[karmic cedar]

I have no GPU

😞

[mortal delta]

well yeah, creative block stinks.

[mortal delta]

i was once there not long ago.

[karmic cedar]

It’s not ideal

[astral goblet]

i have an rtx 4080

[karmic cedar]

That’s a pony

[kindred vigil]

does it crash alot with cuda memory errors and with errors that just say it "errored out" without much more explanation

[astral goblet]

https://tenor.com/view/pony-ride-back-ride-skates-gif-13973117

[karmic cedar]

What size stuff are you trying to generate?

[astral goblet]

cuda? what do you think this is?

oh i got an idea somerandomguy. trogdor . annnnnd go

[mortal delta]

I have both windows and linux mint on one machine, i have to run it on linux to prevent constant crashing and i have to create small amounts with minimal settings.

[mortal delta]

and what does that mean?...

[karmic cedar]

Sounds like a vram issue

[astral goblet]

oh man you're in for a treat https://www.youtube.com/watch?v=TfrAf1a9Qhs

[karmic cedar]

Old school meme

[kindred vigil]

whats mint, like a kind of virtual machine

[karmic cedar]

It’s a Linux distribution

[mortal delta]

your saying i can make little cartoons with ai?

[astral goblet]

if you believe

[karmic cedar]

You can scribble and then use a controlnet to transform it into a dingus

[mortal delta]

I might have to look into that along with other use cases too.

[kindred vigil]

it does a cuda memory error if its too big resolution, and it does the other error frequently for no reason

[astral goblet]

basically, any art you can do with other tools, you can potentially do with ai

[karmic cedar]

And furthermore, anything you can imagine can potentially be syntactically exacted to a similar intellectual precision using an artificial platform.

[trail lion]

I dualboot windows and mint also

[mortal delta]

cool

[trail lion]

amd card, and based on my experience with vast ,I get nowhere near a 3090 with my amd 6800 xt, which is why I use vast

[astral goblet]

precisely

[trail lion]

but the amd card is more than fine for general inference

[karmic cedar]

Which is, pardon my human—fucking frightening, among other feelings.

[astral goblet]

you find you get better performance on rocm in linux than directml in windows?

[mortal delta]

vast?

[astral goblet]

being a creator, that's a daunting task

[mortal delta]

directml always crashes for me on windows

[karmic cedar]

It’s a matter of responsibility and context.

[trail lion]

never tried it, but I only have windows for some one-off things, linux is my main daily driver

[charred mesa]

https://www.reddit.com/r/StableDiffusion/comments/1brpzf7/sd3turbo_is_better_than_midjourney_6_in_both/

why are people only NOW talking about Turbo suddenly

it's been like a week since we've known about Turbo

[astral goblet]

https://www.youtube.com/watch?v=86x-u-tz0MA this chicken soup for the soul lady talks about creativity. might help you cope.

[karmic cedar]

People aren’t all aligned on what it means for something to be “turbo”

[charred mesa]

I mean it's better late than never

[trail lion]

so happy in mint, whenever I have to boot into windows, it's painful

[karmic cedar]

Also I think information fatigue might be setting in

[astral goblet]

anticheat video games right?

i had to keep windows installed for battlefield

[trail lion]

like VR stuff is an example

VR in linux, forget it

[astral goblet]

i would've thoguht valve would hhave steamvr tuned for linux but yeh, i was wrong about that

[karmic cedar]

I have Vision Pro videos I shot for a platform I don’t know if I’ll ever get to experience firsthand (unless I get to 70 and there’s one in a cardboard box somewhere)

[trail lion]

microsoft is killing their vr, so I have another year and then my headset will die

[mortal delta]

oh i just remembered, ive been dabatinig on starting a youtube or a video game creation, or just do something else, but as you can tell those are pretty hard to achieve anything with.

[trail lion]

having said that, 3 body problem released some vr headset that supposedly trumps almost anything on the market. could be interesting to see where that goes. I dont want meta, or apple

[astral goblet]

the 3 body problem headset isn't real lol

[trail lion]

it's in the hands of select few, it's about as real as SD3 😄

[astral goblet]

#🏞｜general-with-images message TROGGDORRRR

[trail lion]

you're legit one of the few I have met that gets that reference. Maybe you're a genx r like me

[astral goblet]

am a proto millenial. right on the cusp of x and m

[karmic cedar]

I buried a strongbad mask once under a rock in Hells Canyon and geotagged it as Nacho Point. ~2006

[astral goblet]

did y'all see the new homestar episode that came out couple months ago?

[trail lion]

burninating the countyside...burninnating the peasants...

[karmic cedar]

No idea there was a new one

[astral goblet]

geo caching is fun

[astral goblet]

came out of NO WHERE.

https://www.youtube.com/watch?v=2-QQpL1YjeU

this came out too a year ago https://youtu.be/xixgDV_9RJI

really happy those guys are still creating

[pearl ocean]

I watched that series!!!

[pearl ocean]

I’m getting back into 2042

[trail lion]

"baby throw" lol

[astral goblet]

it'll be the next book series i get into i think

[pearl ocean]

[astral goblet]

same the new season is a-okay. steam profile is on my discord info if you want to add me and squad up sometime. i'm not as great at it as i used to be though

[broken smelt]

Oh please how great do you have to be to repair a few things and help teammates out

Just kidding, apparently super great because I'd never noticed any of that except for a few folks on my teams the last time I played

[astral goblet]

support is a good class to play

i been upping my engineer game lately. so many good pilots out there

[karmic cedar]

what if you could join those games as just like, a painter or a street cleaner

Battlefield 2042 + Power Wash Simulator

They could even call it something like “BattleWash PowerField 2001”

i’m going silent now

[true canopy]

anyone using forge here? im....having...problems with upscalers

[pearl ocean]

im trying to set it up myself, I don't know how to get the SVD going

[true canopy]

u gotta use 1024x576 res

and have the checkpoint for the svd

and ur good to go

[pearl ocean]

yeah, I just saw a video saying I need to download a check point model

[true canopy]

im not sure there are a sdxl svd, but there is a 1.5 one

https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt

[pearl ocean]

im downlaoding this.
https://civitai.com/models/207992/stable-video-diffusion-svd

not sure if its XL though

[true canopy]

it seem thats the 1.5 one

but mine is almost 9gig so i wonder wth i downloaded

[pearl ocean]

9gigs worth of videos

--

[shut owl]

TakeRep <User:User> [Num:Whole number]

Invalid arguments provided: Not enough arguments passed

[pearl ocean]

my question is, what can I use all these image creation stuff for

[true canopy]

motion bucket id is how much movement u want in the video

if thats what ur asking

[astral goblet]

is actually based on sd2.1

[true canopy]

thats a bit confusing

i tried it with sdxl and it didnt work

[pearl ocean]

it would be nice if Stable Diffusion had something like Sora

would probs need like 128gb of VRAM though

[true canopy]

it wouldent work on local consumer hardware

[astral goblet]

i heard part of sora is a model they trained to animate primatives in unreal engine and then feed them into another model. bit of hand coded components to it. a chain of specialized experts

[pearl ocean]

Is the stable diffusion videos even worth making?
like their only 4 secs long, and the quality isn't really that good, and things look very distorted 😦

[astral goblet]

takes artistic direction and skilled editing to really make anything of any quality.

[karmic cedar]

Not really. I think at the rate SVD has matured compared to some other open source img2vid platforms, it may be soon that something else overtakes it more completely

New fuck-off upscaler has dropped

https://github.com/XPixelGroup/DiffBIR?tab=readme-ov-file

[pearl ocean]

I think I might need to wait a little longer until it gets better

[karmic cedar]

That’s also just my opinion, could be that SVD gets another leap forward but…I think that’s all going to come down to what happens with our primary interest 😊

[quick moat]

Quick question, what’s are y’all’s thoughts on using a Radeon Vega Frontier Edition (16 gb) for stable diffusion?

[karmic cedar]

I mean SD does have AMD support

So it’s really a question of use case and vram

[astral goblet]

i found a site with a ton of free animation clips, if anyone is interested in sauce for the animatediffs https://www.vecteezy.com/video/8661328-animation-infinite-looping-triangle-black-and-white-seamless-loop-motion-background

[karmic cedar]

At this point tho it is a pretty old card

[quick moat]

I’d like to generate the best images possible on a budget but tbh I just herd that the more vram you have the better, idk if there’s any o the buying factors I should take in for it like hrz or core count

[astral goblet]

i used a vega 64 with 8gb when i first started

rocm on arch linux

[quick moat]

Ye I plan to run it on Linux with a little local web server

[astral goblet]

didn't even have sdp support back then

[quick moat]

I’m just debating in like the best gpu to buy (I’m running this in a Mac Pro btw lol)

What factors I should know to make it better

Like speed wise

But I do want the best quality possible and tbh I don’t mind waiting as much which is why the budget is low

250 ish

[astral goblet]

linux, sdp, some kind of tiling , should be good to go

quality isn't really hardware dependent. speed is

[quick moat]

What is sdp again?

I assume stable diffusion but idk what the p means

[astral goblet]

single dot something or other. it's like xformers but is hardware agnostic

oh i was wrong. scaled dot product .. i guess

[pearl ocean]

I just use windows

[broken smelt]

Tbh, hardware isn't really important with the existence of multi diffusion

[astral goblet]

i switched back over to team dows after getting my 4080. why bother with shitty linux nvidia drivers that don't support the premium features ?

[shell tendon]

...not sure i agree with that

[astral goblet]

https://multidiffusion.github.io/pics/fusing.mp4 this example i think is what he means. it can do very large images. provided that one subject isn't spanning all those regions. its not like tiled-diffusion or ultimate upscaler in that regard

[shell tendon]

yeah, maybe i missed something, but "hardware isn't really important" sounded like a bit broad of a statement

[jade mason]

SD3 when?

[shell tendon]

maybe hardware isn't critical for specific things... but it's still important no matter what, unless your time is worthless

[shell tendon]

2024 perhaps

[astral goblet]

its just something poor people say to cope.
||heh||

[shell tendon]

yeah, i can't imagine trying to do this stuff with 8gb vram etc like i hear some people talk about

it was a pain on a 12gb 3080 imo

[foggy halo]

guys sd3 release when

[trim magnet]

in 3 hours